Multi-Modal Foundation · allenai
OlmoEarth v1 Nano
A lightweight vision transformer for remote sensing analysis across Sentinel and Landsat satellite imagery.
What OlmoEarth v1 Nano is
OlmoEarth v1 Nano is a compact Earth-observation foundation model developed by the Allen Institute for AI. Built on a Vision Transformer Nano architecture with 1.4 million parameters, it represents the smallest checkpoint in the OlmoEarth suite.
The model addresses remote sensing tasks across earth science, climate, and conservation. It accepts single scenes and image time series from Sentinel-1, Sentinel-2, and Landsat platforms, providing a minimal computational footprint tailored for embedded devices and high-throughput data pipelines.
How it works
The model employs a ViT-Nano (Vision Transformer) architecture with 1.4 million parameters. It processes multi-modal satellite observations, accepting both radar and optical data from Sentinel-1, Sentinel-2, and Landsat satellites in the form of individual images or multi-temporal image series.
The network can be used to generate image embeddings or serve as a base network for fine-tuning on downstream remote sensing tasks, including image segmentation.
Key facts
- Architecture
- ViT-Nano
- Parameter count
- 1.4M
- Supported sensors
- Sentinel-1, Sentinel-2, Landsat
- Supported input formats
- Single images and image time series
What you can use it for
- Satellite Feature Embedding
- Generating compact numerical representations from Sentinel and Landsat scenes for downstream analytical workflows.
- Remote Sensing Segmentation
- Fine-tuning the model to classify and segment land cover and surface features in satellite imagery.
- Time Series Earth Observation
- Analysing temporal sequences of satellite scenes to observe environmental and planetary changes over time.
- High-Throughput and Embedded Inference
- Running inference on edge devices or in resource-limited computational environments where standard foundation models are too large.
Run it here
We list this model but cannot run it yet: no pipeline_tag — needs a human.
Everything on this page still applies. Running it needs a loader for this repository, not a bigger machine.
Strengths and limits
Good at
- Contains only 1.4 million parameters, enabling high-throughput pipelines and embedded deployment.
- Handles both static single-frame imagery and temporal image sequences.
- Compatible with multiple satellite sensor streams including Sentinel-1, Sentinel-2, and Landsat.
Watch out for
- The small 1.4-million-parameter capacity offers limited representational depth compared to larger foundation models.
- Input imagery is restricted to Sentinel-1, Sentinel-2, and Landsat data streams.
Inputs and outputs
You provide
Sequence or text
A protein or nucleotide sequence, or free text — whichever this model was trained on.
You get back
Mean-pooled representation, one vector per input.
Licence
Check the licence before commercial use · See model card
The See model card licence carries terms worth reading before you depend on it.
Questions
Can I use OlmoEarth v1 Nano commercially?
Check the licence before commercial use. The See model card licence carries terms worth reading before you depend on it.
What satellite instruments does OlmoEarth v1 Nano support?
The model supports radar data from Sentinel-1 as well as optical imagery from Sentinel-2 and Landsat.
Can the model process temporal data?
Yes, it accepts both individual images and multi-frame image time series.
How large is the model?
It uses a ViT-Nano architecture consisting of 1.4 million parameters.
What downstream scientific tasks can be performed?
Researchers can compute visual embeddings or fine-tune the architecture for tasks such as segmentation across conservation, climate, and earth science domains.
Written from OlmoEarth v1 Nano's own documentation (model card). Statements citing a figure that does not appear there are removed before publication. Licence and running details come from our registry.